cover_letter / art_ZCGgNBNMA30

Dear Amplitude Hiring Team, Amplitude sits at the intersection of product intelligence and AI-native workflows — a platform that doesn't just report on what users did, but increasingly helps teams decide what to do next. That mission resonates directly with work I've been doing for the past several years: building agentic systems that close the loop between data, insight, and action. From implementing the OpenClaw multi-agent orchestration framework (gateway protocol, subagent delegation, profile management, session switching) to shipping MCP servers that expose screen-capture and job-search tools to AI coding assistants, I've been building exactly the class of products this role is designed to lead. ## Technical and AI Foundation My AI/ML work spans from first principles to production systems. In 2004 I hand-coded backpropagation through time in C++ for a protein structure prediction system; that work was accepted at NeurIPS 2014. In 2026 I rewrote the same system in PyTorch across five architectures (feedforward, GRU, Transformer, ESM-2, multi-task), scaling from 413 to 8 billion parameters — a 19-million-fold increase — with MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving. On the evaluation side, I built **aeval**, a local-first model evaluation platform covering factuality, reasoning, instruction-following, safety, and code generation. It includes adversarial safety testing with refusal detection, data contamination detection via SHA-256 hashing, bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, saturation detection, and CI/CD regression gates — precisely the kind of probabilistic quality infrastructure Amplitude's JD calls out as a core requirement. I also built an **RL post-training workbench** that benchmarks 12 algorithms (PPO, GRPO, DAPO, DPO, SimPO, and others) across TRL, VeRL, OpenRLHF, and NeMo RL with live SSE metric streaming, cross-tab workflow lineage tracking, and GPU Docker passthrough — giving me deep fluency in the LLM post-training landscape that underpins modern agent behavior. For agentic product delivery specifically, I designed and shipped **OpenClaw**, a multi-agent orchestration framework with a gateway protocol, subagent delegation, profile management, and session switching — coordinating AI workflows across real estate, insurance, health/dental, and financial-markets domains. I built the **Vantage** platform with 40+ agent tools, enforced mutation approvals, and streaming agent chat over SSE, and shipped MCP servers exposing screen-capture and job-search capabilities to external AI coding assistants. These are not prototype integrations; they are production systems with App Store distribution, signed builds, and paying customers. ## Why This Role The Agentic Amplitude team has already shipped a Global Agent, Custom Agents, and an MCP surface — and is now looking to define the next 12–24 months of that strategy. That is exactly the kind of 0-to-1-to-scale arc I've navigated repeatedly, and the combination of MCP expansion, agent eval infrastructure, and customer workflow discovery maps directly onto work I'm doing today. What specifically excites me about this role is the MCP surface area. Amplitude's opportunity to become the analytics layer that coding agents, CLIs, and IDEs can query natively — not just a dashboard users visit — is a genuinely new product category. I've built MCP servers that expose tools to AI coding assistants, and I understand the protocol, the context-management tradeoffs, and the latency constraints involved. Expanding that surface so that a developer's agent can pull Amplitude cohort data or funnel analysis mid-workflow, without leaving their IDE, is a product bet I find compelling and technically tractable. ## Selected Relevant Experience - **OpenClaw multi-agent orchestration framework** — designed and implemented gateway protocol, subagent delegation, profile management, and session switching; coordinating AI agent workflows across multiple industry verticals in production. - **MCP servers for screen capture and job search** — shipped MCP servers exposing tools to AI coding assistants, with App Sandbox compliance, signed mac-arm64 distribution, and MAS-safe embedded video; directly analogous to Amplitude's MCP surface expansion goals. - **aeval evaluation platform** — built end-to-end eval infrastructure (factuality, reasoning, safety, code generation) with statistical rigor (bootstrap CI, Welch's t-test, Cohen's d), CI/CD regression detection, and automated safety gates — the measurement foundation the JD explicitly requires for probabilistic products. - **Vantage agentic platform** — 40+ agent tools, enforced mutation approvals, streaming SSE agent chat, async job polling, idempotent endpoints, and per-role artifact generation; led 0-to-1 product strategy through customer discovery and iterative refinement. - **Intuit ICE platform** — achieved 275% YoY growth in engagements, scaling to 675M+ in FY23; reduced developer onboarding from 2–3 weeks to minutes; scaled throughput from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections with sub-25ms TP99. Demonstrated ability to own a platform product at enterprise scale with measurable adoption outcomes. - **RL post-training workbench** — benchmarked 12 RL algorithms across TRL, VeRL, OpenRLHF, and NeMo RL with live metric streaming, GPU Docker passthrough, and standardized throughput/memory/convergence reporting; reflects the depth of LLM tooling fluency the role requires. - **De Anza College adjunct faculty** — teaching Java, cloud computing, data analytics, and ethical hacking since 2018; reinforces the thought-leadership and communication dimension of the role, translating complex technical concepts for varied audiences. ## Closing Amplitude's stated goal — helping teams analyze, test, and optimize user experiences faster than ever — is a goal I've been building toward from the other side of the table: as a founder shipping AI products, as a platform PM scaling infrastructure to hundreds of millions of engagements, and as a researcher who has cared about measurement and rigor since NeurIPS 2014. The Agentic Amplitude team's bias toward rapid iteration and comfort with evolving technology matches how I operate. I would welcome the opportunity to discuss how my background maps to the specific challenges ahead. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu · 909-731-9011 · felixamoruwa.info